Product Recommendation

Introduction

Product Recommendation allows your AI agent to suggest the right products to users during conversations.

Using your product catalog, tags, and conversational context, the agent can guide users through discovery, comparison, and decision-making—just like a knowledgeable sales assistant.

Product Recommendation is primarily used by Shopping Assistant agents and works across channels such as websites, Shopify storefronts, and voice. 

How Product Recommendation Works

Product recommendations are powered by three key inputs:

1. Your Product Catalog

2. User Intent from the Conversation

3. Context and Rules Defined Through Training

When a user asks a shopping-related question, the agent:

• Understands the user’s intent

• Matches it against your catalog data

• Suggests relevant products based on context

Recommendations are conversational, not static

Prerequisites for Product Recommendation

Before using product recommendations, ensure that:

• You have a Shopping Assistant agent

• Your product catalog is trained using:

  • Shopify sync, or

  • Product APIs

• Products have meaningful tags and attributes

• The agent is tested in the Playground

Without catalog training, product recommendation will not work.

What the Agent Can Recommend

The Shopping Assistant can:

• Suggest products based on user needs

• Narrow down options using follow-up questions

• Compare products or variants

• Recommend alternatives if a product is unavailable

• Respond using real catalog data

The agent does not invent products—it only recommends what exists in your catalog.

Role of Product Tags in Recommendations

Product tags are one of the most important inputs for product recommendation.

Tags help the agent:

• Understand product categories

• Match products to user intent

• Filter and group similar items

Examples of Useful Tags

• Category: running-shoes, formal-wear

• Use case: gym, office, travel

• Attributes: cotton, waterproof, wireless

• Audience: men, women, kids

Well-tagged products lead to more accurate and relevant recommendations. 

Recommendation Behavior During Conversations

The agent uses conversational context to refine recommendations.

For example:

• If a user says “I need shoes for running,” the agent looks for products tagged with running-related tags.

• If the user adds “under ₹5,000,” the agent filters by price.

• If a product is unavailable, the agent suggests alternatives.

The agent adapts recommendations as the conversation evolves.

Using Product Recommendation Across Channels

Product recommendations work consistently across channels, including:

• Website chat widgets

• Shopify storefronts

• Voice interactions

• Playground testing

The recommendation logic stays the same; only the presentation differs by channel.

Testing Product Recommendations

Before going live, always test recommendations.

To test:

1. Open Test AI Agent

2. Ask product-related questions

3. Try different intents and constraints

4. Validate suggested products against your catalog

Testing helps ensure:

• Correct products are suggested

• Tags are working as expected

• Responses are clear and helpful

Improving Recommendations with Instant Retrain

If a recommendation is incorrect or not ideal:

• Use Instant Retrain in the Playground or Conversations

• Correct the response

• The correction is saved as Text training

This helps refine recommendation behavior over time.

Best Practices for Product Recommendation

• Keep product data accurate and up to date

• Use clear and consistent tags

• Avoid duplicate or overlapping tags

• Test recommendations regularly

• Combine catalog training with text rules for edge cases

Good data leads to good recommendations.

Common Mistakes to Avoid

• Missing or vague product tags

• Expecting recommendations without catalog training

• Using inconsistent naming conventions

• Not testing after catalog updates

If the catalog isn’t well-structured, recommendations will suffer.

When to Review Product Recommendation Setup

You should review recommendations when:

• New products are added

• Product tags change

• Users report irrelevant suggestions

• Business priorities change (e.g., new categories)

Small adjustments can significantly improve outcomes.

What’s Next?

Once product recommendations are working well, the next step is to monitor real user conversations and step in when needed.

 Next User Guides: Conversations & Human Takeover